FORECAST.LINEAR in Excel: Fitted Values, Extrapolation and Paired Inputs

• Introduced: Microsoft documents FORECAST.LINEAR replacing FORECAST in Excel 2016; the function catalogue also marks the new identity as a 2016 addition. An exact first build or preview date is not established.
• Availability: official help lists Excel for Microsoft 365, Excel for Microsoft 365 for Mac, Excel 2024, Excel 2024 for Mac, Excel 2021, Excel 2021 for Mac, Excel 2019, Excel 2016.
• Checked: October 14, 2026; official FORECAST.LINEAR reference. Documentation and independent models; no native Excel execution or wider platform certification.

=FORECAST.LINEAR(x,known_ys,known_xs)

All three arguments are required. The target X comes first, followed by observed Y and then observed X. Microsoft documents the equation a+b*x, with slope and intercept calculated from the paired data.

Microsoft introduced this identity as the replacement for FORECAST in Excel 2016. The older spelling remains available for compatibility; a recommendation to use the newer name does not make every forecasting-related function interchangeable.

The formula fits a straight line. It does not discover seasonal structure or return a prediction interval.

Enter X 0,1,2,3 in A2:A5 and Y 10,13,12,17 in B2:B5:

=FORECAST.LINEAR(4,B2:B5,A2:A5)

Expected result: 18. Mean X is 1.5 and mean Y is 13. Centered paired products total 10, while centered X squares total 5. Slope is 2 and intercept is 13-2*1.5=10, so the fitted line is Y=10+2*X.

The last observation is 17 at X=3, but the line’s fitted value there is 16. The function does not simply add the last observed change to the last observation.

=FORECAST.LINEAR(1.5,B2:B5,A2:A5)

Expected result: 13. This target lies inside the observed X interval. It also equals mean X, and the intercept-containing fitted line passes through the pair of sample means.

At X=1 the fitted value is 12, while the observed Y is 13. Even at an observed X, a least-squares line generally does not reproduce every noisy observation exactly.

INTERCEPT and SLOPE provide separate coefficient checks. Using their expected values, 10+2*1.5 reproduces 13.

Put targets -1,0,1.5,4,6 in D2:D6. In E2 enter:

=FORECAST.LINEAR(D2,$B$2:$B$5,$A$2:$A$5)

Fill down. Expected fitted values are 8,10,13,18,22. The absolute references retain the same training data while the target changes.

Targets -1,4 and 6 lie outside the original interval 0 through 3. Label them as extrapolations; the straight line continues mathematically, but an underlying process may not continue in that way.

Use a separate table with X 0,2,5,9 and Y 10,14,20,28. These values follow Y=10+2*X exactly.

At target X=6, expected FORECAST.LINEAR is 22. Substituting row positions 1,2,3,4 for those X values would fit a different model. A row number is not automatically the correct time, distance or exposure variable.

Keep actual independent-variable values and their units. If X represents elapsed time, define its origin and scale rather than relying on a display label.

For the original fixture, add 100 to all observed X values and use target 104. The slope stays 2 and the intercept becomes -190. Expected result remains 18.

Keeping target 4 while shifting only the observed X values would instead evaluate a different, distant X and give -182. Transform the target and observations consistently when changing coordinates.

Microsoft specifies #VALUE! for nonnumeric target X, #N/A for an empty observed series or unequal data-point counts, and #DIV/0! when observed X variance is zero.

The dedicated reference does not give a complete mixed-type observation matrix. All worked data here are numeric and paired. Missing records, error-containing ranges and extreme magnitudes need native testing. A successful numeric output is a fitted value, not evidence that the data quality or forecasting assumptions are sound.

• Microsoft FORECAST.LINEAR reference.
• Editorial testing standards.

Expected outputs are independent mathematical calculations or explicitly documented rules, not captured native Excel results. Examples use fictional data, English function names and comma separators. Native floating-point boundaries remain untested.